3.1 A conjecture: Welfare institutions mediate the effect of egalitarian values on social cohesion
Theory and previous findings suggest that institutions and policies mediate the effect of values on social cohesion. In this section, we present a first preliminary exploration of this triangular relationship. This exploration is limited in focus and scope. First, of those effects that values may have on social cohesion, we choose to focus on the extrinsic direct and indirect effect, and leave aside the intrinsic effect of shared values (Figure 1). Second, to keep the empirical exploration in scope, we choose to focus on only one dimension of social cohesion, namely trust. We choose trust, as this is one attribute of social cohesion for which the effect of institutions is well studied and robust (see above). Finally, as we argue further below, we limit our focus on the particularly relevant nexus of one type of values, namely egalitarian values, with one particular policy regime, namely welfare policies. Extending and deepening our exploration will be a task for future efforts.
We define welfare policies as policies that aim at providing a country’s population with a secured minimum living standard over a range of different life stages and conditions. A welfare state regime, or, interchangeably, welfare policy regime, is the network of the social institutions that create welfare policies. A welfare policy regime encompasses, for instance, educational policies, health care schemes, retirement schemes as well as unemployment and poverty programmes. Today’s differences in welfare policy regimes evolved historically,
and differ primarily in terms of their scope, or degree of eligibility and inclusion (Esping-Andersen, 1990). The liberal type of welfare policy regimes is at one end of the spectrum.
It evolved from the European poor-relief policies of the 19th century, and focuses on means-tested social support. In these welfare policy regimes, entitlement to a social service is strictly dependent on need, which is assessed according to fixed rules. Typical examples are the welfare policy regimes of the UK or Australia. At the other end of the spectrum lies the social democratic, or universalistic, regime type. It evolved as a response of socialist parties to the liberal welfare regime type at the turn of the 19th to the 20th century and builds upon the notion of universal eligibility (Esping-Andersen, 1990). Here, welfare policies that benefit the entire population without prior means-testing dominate. A typical example of such policies are universal child care policies (Rothstein, 2008). The Scandinavian countries provide prominent examples for universalistic welfare regimes (Vrooman, 2012).
Welfare institutions are important factors for social cohesion in general, and for trust in particular. The literature is consistent on the effects of welfare institutions and points out how important they are in generating and maintaining social cohesion (Edlund & Lindh, 2017; Rothstein, 2001). According to the literature, both types of welfare institutions have distinct effects on trust in a society. Universalistic social welfare institutions are conducive to trust by treating everyone according to the same standards, while means-tested institutions risk increasing perceived social distinctions by emphasising inequality, thus undermining trust (Edlund & Lindh, 2015; Rothstein & Uslaner, 2005). First, concerning universalistic welfare policies, Brewer, Oh and Sharma (2014) show that a higher total social welfare expenditure, which is known to be correlated with a higher proportion of universalistic welfare policies, is associated with greater trust, even when controlling for reversed causality. Furthermore, exploiting the health care reform in the US as an event in the course of which the US welfare policy regime became more universalistic, Mewes and Giordano (2017) show that universalistic welfare policies mitigate factors that usually explain lower social trust. Second, compared to universalistic welfare policies, liberal means-tested welfare policies tend to emphasise social distinctions by stigmatising beneficiaries (Esping-Andersen, 1990; Stuber & Schlesinger, 2006). More recent research shows that means-tested welfare policies do not automatically lead to a “welfare stigma”
(Moffitt, 1983), but are conditional on societal moralistic perceptions on who the
‘deserving’ and the ‘undeserving poor’ are (Calnitsky, 2016). Although the stigmatising effect of means-tested welfare policies is conditional in this way, they may nonetheless increase social tensions. In addition, means-tested welfare policies may increase a sense of social alienation among beneficiaries by violating their personal integrity through the necessary intrusive procedures involved in assessing their economic and social situations (Calnitsky, 2016; Rothstein, 2008).
Having characterised the general association between different types of welfare policy regimes and trust, the next question concerns which role values play for the scope of welfare policies. A minor strand of the literature on welfare policies has begun to focus on the role of cultural factors for the support of different policy regimes, but has not yet made the connection to value concepts from social psychology (Van Oorschot, 2007; Vrooman, 2012). Among these values, Schwartz’ egalitarian values seem to be most in line with universalistic welfare policies. They “socialize people to internalize a commitment to cooperate, to feel concern for the welfare of all, and to act voluntarily to benefit others.
Important values in such cultures include equality, social justice, responsibility, help, and honesty” (Schwartz, 2014a, p. 551). In theory, it should thus be expected that societies with
a strong emphasis on egalitarian values are more likely to create and maintain universalistic welfare institutions. Indeed, individuals in countries where egalitarian as opposed to hierarchical values are emphasised, more strongly support redistribution through welfare institutions. The same effect holds for embeddedness in contrast to autonomy values. This suggests that individuals in egalitarian societies are more supportive of universal welfare institutions, while societies with an emphasis on embeddedness values are more supportive of means-tested institutions (Arikan & Bloom, 2015).
Taken together, although the evidence is still slim, theory and already extant empirics provide for the hypothesis that egalitarian values increase the likelihood that a society adopts a welfare policy regime in which universalistic programmes dominate, which in turn increases both social and institutional trust. Figure 2 illustrates the conjectured relationships in a path diagram. It should be carefully noted that the causal relationship between values, welfare policies and trust can run in either direction. Indeed, this is still an unsolved issue in the literature on welfare policies (e.g. Bergh & Bjørnskov, 2014). We discuss this issue further in more detail below. In the next sections, we explore whether the conjectured relationships between egalitarian values and universalistic welfare institutions on the one hand, and between universalistic welfare institutions and trust on the other hand hold.
Lastly, we also assess whether a universal scope of welfare institutions takes on a mediating role in the relationship between egalitarian values and trust.
Figure 2: Conjectured relationship between egalitarian values, welfare institutions and trust
Source: Authors
3.2 Data and operationalisation
In order to explore empirically the triangular relationship between values, institutions and social cohesion, we drew on cross-sectional secondary data from large-scale surveys, indexes from expert ratings, as well as national statistics. To account for the macro-societal nature of some of the main concepts employed in this research (e.g. cultural values, welfare
institutions), all individual data were aggregated at a country level. In total, data on all variables was available for 63 countries across all inhabited continents of the world. An overview of the countries can be found in the Appendix.
In our empirical exploration, we chose to examine two potentially important country-level antecedents of social cohesion: the countries’ cultural value emphasis on egalitarianism and the ratio of universalistic to means-tested social welfare institutions. In order to keep confounding effects as small as possible, we included multiple control variables, including the pervasiveness of political corruption, level of democracy and economic development.
Political corruption and democracy have been found to be strong determinants of social and institutional trust (Rothstein & Uslaner, 2005) and were thus included as control variables.
We also controlled for educational equality and freedom of discussion, as social trust is likely to be fostered in environments where information can flow freely (Fisman & Khanna, 1999), and where equality of opportunity in areas such as education prevails (Rothstein &
Uslaner, 2005). Since ethnic heterogeneity might influence the sense of belonging to a larger social order (Licht et al., 2007), it was also integrated as a control variable. Lastly, we added economic development and government revenue to rule out differences between countries in terms of economic strength, which is usually associated with a more benign social policy framework and affects how extensively governments engage in distributive expenditures (Muinelo-Gallo & Roca-Sagalés, 2011)5. An overview of all variables and their measurement is provided in the Appendix.
We retrieved data on egalitarianism from the Schwartz Value Survey (SVS)6. In the SVS, participants are asked to rate the importance of different values as guiding principles for their lives on a Likert scale ranging from -1 (opposed to my values), 0 (not important) to 7 (of supreme importance). Importance ratings on six values (equality, social justice, loyalty, honesty, welfare and responsibility) are used for determining a country’s cultural value emphasis on egalitarianism (Schwartz, 2014a). Since individuals and cultural groups might use the value scales differently, the individual-level raw scores were corrected for scale use before aggregation for the cross-country analyses. In this we followed the instructions provided in the SVS.
Recall that our mediating variable is the ratio of universalistic to means-tested social welfare policies. For this, we used the means-tested vs. universalistic policy indicator from the database Varieties of Democracy (Coppedge et al., 2019), which measures the quality of the welfare state on an interval scale ranging from 0 to 5. A score of 0 implies that there are no, or extremely limited, welfare state policies, whereas a score of 5 means that almost all welfare state policies are universal in character (V-Dem, 2019). More fine-scaled indicators on social welfare programmes are rare and would have necessitated a major data collection effort. Because of this and due to its broad coverage in terms of countries and time, this indicator was deemed most suitable for the purposes of our explorative analysis.
5 Intercorrelations between the covariates ranged from -0.73 to 0.79. However, in all cases, multicollinearity did not occur (VIF< 10).
6 The SVS data set was compiled between 1985 and 2005 by Schwartz and over 100 collaborators worldwide. The data set was accessed via the Israel Social Sciences Data Center (Israel Social Sciences Data Center, 2019).
With regard to the outcome variables, social trust was measured by the binary ‘most-people’-question (“Generally speaking, would you say that most people can be trusted or you can’t be too careful in dealing with people?”). This question has been included as an item in all waves of the World Value Survey (WVS), as well as other regional surveys7. In their review of trust measurement, Bauer and Freitag (2018) conclude that the dichotomous most-people measure is still mainly used to assess social trust in spite of several shortcomings. These include limited scale length and interpersonal comparability, as people from different cultures may interpret the most-people question differently. Since more nuanced data are not available for a broad range of countries, we decided to stick with the binary most-people question to approximate countries’ levels of social trust. Reversing the scores from the WVS and European Value Survey (EVS), ensured that for all countries a higher social trust score indicates that people trust each other more.
In order to gauge institutional trust at the country level, we calculated an additive individual-level index based on the measurement of trust in the police, the courts and the parliament on a Likert scale from 1 to 4, which was then aggregated at the national level as a simple average. The scores of the WVS and EVS were reversed, so that for all countries a higher institutional trust index translates into more trust in the respective institutions.
Availability of the data posed a major challenge. The SVS data are available only until 2005, and the WVS and EVS often provided no data earlier than 2005. Hence, checking for the time difference between the SVS data and the data from the other surveys was especially important. Even though cultural values are thought to be quite stable (Schwartz, 2011), some authors found them to change inter-generationally (Inglehart & Abramson, 1994).
Therefore, we compiled data only for those countries where the data collection of all surveys took place within a one-generation period of approximately 20 years, that is to say 10 years before and 10 years after the cultural value measurement. Countries to which this did not apply (Ethiopia, Indonesia, Singapore and Zimbabwe) were given an additional dummy variable in order to control for periods between data collection exceeding a one-generational time span. The analysis revealed that, in fact, the longer periods between data collection have no significant effect on the outcome variables social trust and institutional trust. Hence, there is no statistical support for excluding the four countries mentioned above from the analyses. They were thus kept in the sample for all analyses.
3.3 Methodology
For our quantitative exploration, we employed two different types of data analysis strategies: multiple regression and mediation analyses.
First, we used multiple regression analyses to indicate the strength and direction of the relationship between egalitarian values and the proportion of universalistic vs. means-tested social welfare programmes. Similarly, we ran regressions to test the relationships between the proportion of universalistic vs. means-tested social welfare programmes and social trust (see Figure 2: upper path), as well as institutional trust (see Figure 2: lower path).
7 These include the Afrobarometer, Arab Barometer, European Value Survey and the Latinobarómetro.
Second, we ran mediation analyses to further investigate the triangular relationship between values, institutions and trust. A mediation model seeks to identify and explain the mechanism underlying the relationship between an independent and a dependent variable via the inclusion of a third variable, the so-called mediator (Hayes, 2009). The utility of mediation analyses8 stems from their ability to clarify the nature of the relationship between an independent and dependent variable by assessing the extent to which the effect of the independent on the dependent variable is direct or indirect through the mediator (Iacobucci et al., 2007). In their review of statistical methods for mediation analysis, Iacobucci et al.
(2007) underline the superiority of Structural Equation Modelling (SEM) for determining mediation effects. SEM makes it possible to examine how well a process model linking a predictor variable to an outcome variable through one or more intervening pathways fits the observed data (Hayes, 2009). In a traditional mediation analysis, the model would be fit by a series of linear regression analyses as outlined in Baron & Kenny (1986). The advantage of using SEM is that it fits a single model and provides estimations of the indirect and total effects. However, it should be noted that a non-correlation of the variables’ error terms is assumed when using SEM as a tool for mediation analyses.
Before we proceed to the interpretation of our results, we want to issue a cautionary note on statistical analysis. Statistical analysis testing has recently been criticised in some parts of the social science literature. Within psychology, for example, this debate has been prominent, and was accompanied by a general ‘replication crisis’ (for an overview, see:
Earp and Trafimow (2015); Aarts and Lin (2015)). The major thrust of the critique on statistical significance and p-values is that they are falsely understood as reporting the odds at which a statistical result is true, whereas they actually report “the probability of obtaining the results in hand, assuming that the statistical null hypothesis is true in the population”
(Lambdin, 2012, p. 74). One consequence of this is, that p-values are often presented in a definitive manner even for non-randomly drawn samples or for whole-population samples.
However, for such samples the meaning of statistical significance is restricted, as the first case does not allow inferences on the entire population and the obtained coefficients are already those of the entire population in the second case. Some therefore argue that statistical significance testing should not be done for whole-population samples or non-random samples (Figueiredo Filho et al., 2013; Frick, 1998). Others, like Gigerenzer, Krauss and Vitouch (2004) propose a more informed and transparent use and presentation of statistical significance. We want to stress that we also do not have a randomly taken sample, but run our explorative analysis based on the totality of the available data. We further discuss the nature of our data below in a section on the limitations of our analysis. For now, we want to stress that the interpretation of our results refers to our sampled countries, and inferences beyond our availability-sample are indicative at best. For maximum transparency, we nevertheless follow Gigerenzer et al.’s (2004) best-practice suggestion and present exact p-values accompanied by confidence intervals.
8 For a more comprehensive discussion of mediation analyses and methodological considerations, see Iacobucci, Saldanha and Deng (2007).
3.4 Results
The multiple regression analyses revealed that egalitarian values positively relate to the proportion of universalistic vs. means-tested social welfare programmes, even if other explanatory variables are taken into account (β=0.235, p=0.111, see Table 2). In other words, with every increase of one standard deviation in the cultural value emphasis on egalitarianism, the proportion of universalistic vs. means-tested social welfare programmes increases by almost a quarter standard deviation if the other variables are held constant.
Surprisingly, the multiple regression analysis revealed that out of all variables, only freedom of discussion is statistically significantly linked to the proportion of universalistic vs.
means-tested social welfare programmes (β=0.53, p=0.013). This implies that countries with more freedom of discussion have a larger proportion of universalistic vs. means-tested social welfare programmes.
With regard to the relationship between the scope of social welfare programmes and institutional trust, the results of the multiple regression confirm that the proportion of universalistic vs. means-tested social welfare programmes is significantly positively related to institutional trust (β=0.296, p=0.021, see Table 3). This implies that countries with a larger proportion of universalistic social welfare programmes experience significantly higher levels of institutional trust even if other explanatory factors are controlled for.
Additionally, the results reveal significant negative relations between institutional trust and political corruption (β=-0.738, p<0.01) as well as electoral democracy (β=-0.6, p=0.014).
Table 2: Determinants of universalistic vs. means-tested social welfare programmes Universalistic vs.
means-tested social welfare programmes
Coef. Std. Err. 95% Conf.
Interval P>|t| β
Egalitarianism 0.498 0.307 [-0.118, 1.114] 0.111 0.235
Educational equality 0.286 0.144 [-0.001, 0.574] 0.051 0.321
Freedom of
discussion 2.221 0.86 [0.497, 3.945] 0.013* 0.53
Political corruption 0.447 0.653 [-0.863, 1.756] 0.497 0.143
Electoral democracy -0.981 1.026 [-3.038, 1.076] 0.343 -0.241
Ethnic heterogeneity 0.077 0.428 [-0.781, 0.934] 0.858 0.023
Economic
development 0.000 0.000 [-0.001, 0.001] 0.904 0.021
Government revenue 1.202 1.479 [-1.763, 4.168] 0.420 0.112
Note: N=63. R²=0.428. F(8,54)=5.04. *p<0.05, **p<0.01, ***p<0.001. The dependent variable is universalistic vs.
means-tested social welfare programmes.
Source: Authors
These results suggest that countries with a high pervasiveness of political corruption and high levels of electoral democracy, experience lower levels of institutional trust.
Contrary to the statistically significant positive association between the scope of social welfare programmes and institutional trust, the multiple regression analysis showed that a country’s proportion of universalistic vs. means-tested social welfare programmes is not connected to its degree of social trust (β=0.083, p=0.533, see Table 4). In other words, if all covariates are held constant, the social trust level rises by merely one tenth standard deviation with every increase of one standard deviation in the proportion of universalistic vs. means-tested social welfare programmes. Hence, the results of the multiple regression suggest that countries with more universalistic social welfare programmes do not experience either statistically significantly or substantially higher levels of social trust. In fact, the analysis revealed that out of all variables, only electoral democracy (β=-0.368, p=0.14), government revenue (β=0.345, p=0.017), GDP per capita (β=0.316, p=0.054), and freedom
Contrary to the statistically significant positive association between the scope of social welfare programmes and institutional trust, the multiple regression analysis showed that a country’s proportion of universalistic vs. means-tested social welfare programmes is not connected to its degree of social trust (β=0.083, p=0.533, see Table 4). In other words, if all covariates are held constant, the social trust level rises by merely one tenth standard deviation with every increase of one standard deviation in the proportion of universalistic vs. means-tested social welfare programmes. Hence, the results of the multiple regression suggest that countries with more universalistic social welfare programmes do not experience either statistically significantly or substantially higher levels of social trust. In fact, the analysis revealed that out of all variables, only electoral democracy (β=-0.368, p=0.14), government revenue (β=0.345, p=0.017), GDP per capita (β=0.316, p=0.054), and freedom